Abhishek --

Could you send the full output from your mtt client run with the --verbose flag 

If you'd prefer not to send it to the public list, send it directly to me and 
Josh Hursey (IBM).


> On Jun 21, 2016, at 6:48 AM, Jeff Squyres (jsquyres) <jsquy...@cisco.com> 
> wrote:
> Greetings Abhishek.
> You sent me your INI file in another email thread.
> In general, you need to run all the 5 phases.  During the MPI install, for 
> example, even if you have an "already installed" MPI (i.e., using the MPI Get 
> module "AlreadyInstalled"), you still have to run that phase so that a) the 
> appropriate data structures are created in MTT, and b) the MTT client 
> registers the result of that "install" with the server (and gets an ID back 
> from the server).
> Specifically: even though the MPI install step may effectively be a no-op 
> (because you're using an MPI that is already installed), MTT still needs to 
> report back the "success" of the MPI install step to the server so that it 
> can get an MPI Install ID back from the server to use with subsequent phases. 
>  That's essentially what the AlreadyInstalled module does.
> That being said, I have a few suggestions about the .ini file you sent me:
> - I see you're using the OMPI_Snapshot MPI Get module.  That's fine (and 
> that's what we all do for nightly testing).  But I thought you said you were 
> using your own sources...?  That's why I mentioned the AlreadyInstalled 
> module, above.
> - ...ah, perhaps you meant that you were using your own sources via the 
> http://localhost/... URL you provided via the ompi_snapshot_url?  I'm 
> assuming that this is just for testing purposes, and you'll replace it with 
> the real nightly URL, later.
> - Note that in the MPI install section, the templated supplies "-j 32" as 
> "make" arguments.  This does a 32-way parallel build.  This is probably only 
> recommended for 16-core machines (or more).  My point: adjust that -j value 
> to suit your local machine.
> - Do you only care about testing the TCP BTL?  The MPI Details section 
> contains the "mca" value, which is currently only set to test the TCP 
> transport for MPI.
> - I see you encoded the github password for the ompi-tests repo in your .ini 
> file.  This is fine, but as you probably figured out, but the drawback is 
> that you can't email your .ini file to a public list (e.g., this one) for 
> discussion.  If you care, as an alternative, I *think* you can stash this git 
> username/password on your local filesystem, instead.
> - Similar to the above, I see your MTT database password in your .ini file.  
> FWIW, I keep my Cisco MTT database password in a separate file on the 
> filesystem, and access it in my .ini file like this:
> mttdatabase_password = &stringify(&cat("/home/mpiteam/mtt-db-password.txt"))
> This allows me to save my MTT .ini file in git / share it with others, and 
> not compromise my password.
>> On Jun 20, 2016, at 2:28 PM, Abhishek Joshi <abhishek.jo...@broadcom.com> 
>> wrote:
>> While trying to submit results for MTT , I get the following error:
>> Invalid mpi_install_id () given. (Not provided). 
>> I am using the following cmd:
>> ./client/mtt   --trial --file ./ompi-core-template.ini_sample --print-time 
>> Also, I am using a custom build, i.e MPI sources are my own.
>> -- 
>> Abhishek
>> _______________________________________________
>> mtt-users mailing list
>> mtt-us...@open-mpi.org
>> Subscription: https://www.open-mpi.org/mailman/listinfo.cgi/mtt-users
>> Link to this post: 
>> http://www.open-mpi.org/community/lists/mtt-users/2016/06/0875.php
> -- 
> Jeff Squyres
> jsquy...@cisco.com
> For corporate legal information go to: 
> http://www.cisco.com/web/about/doing_business/legal/cri/

Jeff Squyres
For corporate legal information go to: 

Reply via email to